Mixed models and ensemble methods and their performance in LGD modeling.
|Marek J. Karwanski , Urszula Grzybowska|
Szkoła Główna Gospodarstwa Wiejskiego (SGGW), Nowoursynowska 166, Warszawa 02-787, Poland
Loss Given Default (LGD) plays an essential role in credit risk modeling. LGD measures the credit loss it can be expressed as a percentage of exposure, which is lost in the case of an event when the borrower does not fulfill his obligations (default). It is treated as a random variable with bimodal distribution.
Many methods for LGD are proposed in literature. Regression-type models using appropriate explanatory variables, named risk factors, are generally preferred as the most flexible approach. For LGD estimation advanced statistical models such as beta regression can be applied. The parametric methods, however, require amendments of the “inflation” type that lead to mixed modeling approach. Contrary to classical statistical methods based on probability distribution, ensemble models such as gradient boosting or random forests operate with information and allow for more flexible model adjustment. The aim of the research is to present results of LGD modeling using mixed models and compare it with ensemble approach.
Calculations were done on real data from one of the biggest Polish bank. They related to the overdrafts in the segment of Small and Medium Enterprises (SME). The overdrafts were chosen because revolving loans don’t depend on eligible collaterals or guarantees and the risk mitigation effect can be calculated using a consistent methodology which allows for pure comparison of different models.
Presentation: Poster at 8 Ogólnopolskie Sympozjum "Fizyka w Ekonomii i Naukach Społecznych", by Marek J. Karwanski
See On-line Journal of 8 Ogólnopolskie Sympozjum "Fizyka w Ekonomii i Naukach Społecznych"
Submitted: 2015-09-07 11:21 Revised: 2015-09-07 11:21